OCI AI Foundations Associate (1Z0-1122-25) Flashcards¶

60 minutes

40

65%

USD 245 list price; Oracle frequently runs free certification periods for its AI and OCI foundations exams

Oracle refreshes these annually with a year suffix in the exam code

Oracle CertView, online proctored

predict a category. Spam or not spam, which of five defect types, will the customer churn

predict a continuous value. House price, tomorrow's demand, time to failure

identify points that do not fit the learned pattern

the system does not produce discriminatory outcomes across groups

users know they are interacting with AI and understand its role

decisions can be explained to the people they affect

a human is responsible for outcomes

personal data is protected throughout the lifecycle

the system behaves reliably, including under unexpected input

systems performing tasks that would otherwise require human intelligence

systems that learn patterns from data rather than following explicit rules

machine learning using multi-layer neural networks

learning from labeled data to predict a label

finding structure in unlabeled data

learning a policy from reward feedback in an environment

grouping similar items without predefined labels

creating the input variables a model learns from

the data used to fit the model

the data used to tune hyperparameters

held-out data used once for an honest performance estimate

a model that memorizes training data and generalizes poorly

a model too simple to capture the underlying pattern

the tension between a model being too simple and too sensitive to its training data

the table of true and false positives and negatives used to evaluate a classifier

the proportion of positive predictions that were correct

the proportion of actual positives that were correctly identified

the harmonic mean of precision and recall

degradation of model performance as live data diverges from training data

receives the features

learn intermediate representations. "Deep" means more than one

produces the prediction

the basic unit computing a weighted sum of inputs passed through an activation function

a learned parameter scaling an input to a neuron

a learned parameter offsetting a neuron's weighted sum

a non-linear function applied to a neuron's output, enabling the network to learn non-linear patterns

the rectified linear unit, the common default activation for hidden layers

the output activation producing a probability distribution across classes

passing inputs through the network to produce a prediction

the measure of how far a prediction is from the correct answer

the algorithm computing each weight's contribution to the loss

the optimization method adjusting weights to reduce loss

a value learned during training, such as a weight or bias

a value set before training, such as learning rate or batch size

the step size used when updating weights

one complete pass through the training dataset

the number of examples processed before each weight update

randomly disabling neurons during training to reduce overfitting

a convolutional neural network, suited to images and spatial data

a recurrent neural network, suited to sequential data

a recurrent architecture using gates to handle longer sequences

the attention-based architecture underlying modern language models

a generative adversarial network, training a generator against a discriminator

a network learning to reconstruct its input through a compressed representation

low-latency high-throughput interconnect used for distributed GPU training

instruction only, no examples

include a handful of worked examples in the prompt

ask the model to reason step by step before answering, which improves multi-step tasks

instructions setting role, tone, and constraints for the whole conversation

supply authoritative source text in the prompt and instruct the model to answer only from it

require the answer to reference the supplied sources so a reader can check

instruct the model to say it does not know when the context lacks the answer

a model that produces new content rather than only classifying existing content

a model that distinguishes between classes rather than generating data

a generative model trained on large text corpora to predict the next token

the unit of text a model processes, roughly a word or word fragment

a vector representation of text where semantic similarity corresponds to closeness

the mechanism weighing the relevance of other tokens when processing each token

the parameter controlling randomness in token selection during generation

the maximum number of tokens a model can consider in a single request

the initial large-scale training that teaches a model language

additional training on task-specific data to change model behavior

fine-tuning specifically to make a model follow instructions

reinforcement learning from human feedback, used to align model outputs to preferences

a parameter-efficient fine-tuning method training a small set of additional weights

prompting with an instruction and no examples

prompting with a small number of worked examples

asking the model to reason step by step before answering

confident but incorrect model output

retrieval-augmented generation, retrieving relevant text into the prompt before generating

a store for embeddings supporting similarity search

Oracle Autonomous Database's native vector storage and similarity search capability

an LLM operating in a loop with tools to complete a multi-step task

Custom models for classification and entity recognition

managed JupyterLab with GPU options

versioned model storage with metadata and provenance

models served as HTTP endpoints with autoscaling

repeatable training or processing runs

orchestrated multi-step ML workflows

deploy, fine-tune, and evaluate foundation models without writing serving code

Object Storage for datasets, File Storage for shared access during training, Block Volumes for attached performance

the ready-made service for sentiment, entities, key phrases, translation, and PII detection

the ready-made service converting speech to text

the ready-made service for image classification, object detection, and OCR

the service extracting structured data from forms and documents

the managed service providing foundation model inference, embeddings, and fine-tuning

isolated capacity for predictable generative AI throughput

the managed service providing retrieval-augmented agents over enterprise data

Autonomous Database's native vector storage and similarity search

the capability translating natural language questions into SQL against the database

the platform for building, training, deploying, and monitoring custom models

the service for creating labeled datasets for supervised training

a compute shape providing direct access to GPUs without virtualization overhead